Agentic Operations Kit
A versionable knowledge core that translates the entire article base into agentic work contracts — with platform-specific adapters instead of six diverging tool manuals.
The value is not a "master prompt." It lies in reusable work contracts: every task gets an objective, sources, allowed actions, output schema, acceptance criteria, handoff format and explicit stop signals.
One canonical core, many adapters
Five operational levels
| Level | Capability | Result |
|---|---|---|
| Assignment | Cleanly separate objective, context, intent and boundaries | Project brief and task contract instead of a vague ask |
| Context | Deliberately steer sources, handoffs and work states | Context manifest, source map and handoff protocol |
| Quality | Test results, deploy counter-models, make expertise visible | Acceptance criteria, review loops and eval sets |
| Agents | Shape roles, tools, autonomy and stop signals | Agent harness with gatekeeper and audit trail |
| Sovereignty | Choose data zones, local options and safe execution | Security profile, tool allowlist and approval gates |
Seven packages, one core
Task Contract
objective: What verifiable result should emerge?
context: Which facts, files and decisions apply?
sources: Which sources are binding, which supplementary?
allowed_actions: What may the agent read, change, run, send?
not_allowed: Which boundaries apply explicitly?
delivery: Files, structure, language, style and location
acceptance: How is "done" objectively recognizable?
stop: When must the agent halt and escalate?
handoff: Which decision, remaining work and evidence are handed over?What explicitly does not happen
- ✕No blanket super-prompt mixing all tasks, platforms and permissions.
- ✕The full knowledge library is not loaded into every session automatically.
- ✕Third-party skills and MCP servers are not adopted without provenance, rights and test review.
- ✕Every automated external contact, file write or credential access has a visible permission and stop rule.
- ✕Every flow produces a handoff with result, evidence, open decisions and next step.
- ✕The vault stays readable without special software: plain Markdown is always the fallback interface.
Recommendation
The starter vault contains the binding work contracts (house rules, task contract, agent harness, evaluation, handoff) in three languages. Extend it with the full article library and native adapters once the large folders are available as ZIP.
The seven packages in detail
Obsidian Start Vault
A complete, versionable knowledge core for an immediate start — the editorial source of all further packages.
- →Core rules, workflows and recipes
- →Work contracts: house rules, task contract, agent harness, evaluation, handoff
- →Start path (15-minute start) and vault index
- →Language versions DE · EN · TR
The ZIP opens directly as a vault and a new task can be set up as project brief, task contract and handoff in 15 minutes.
IDE Operations Kit
Reliable rules and workflows for coding agents — the biggest direct lever for daily agentic work.
- →One canonical rule file plus native formats per tool
- →Project brief, commit/test workflow, handoff templates
- →Short and reference-oriented so the context window does not overflow
Every rule file points to the same house rules, contains no secrets, and the example flow plan → change → test → handoff works.
OpenClaw Operations Kit
Safe, message-capable agent workflows and extensions for a self-hosted gateway.
- →Workspace template and channel-routing model
- →Session/handoff rules, tool and credential policy
- →Clear separation of intake and executing actions
New extensions have a traceable review, test, approval and rollback path; unconfirmed messages cannot trigger external actions.
Hermes Operations Kit
Use persistent learning and reusable skills under control.
- →Rules for which memories may become permanent
- →How a run becomes a versioned, reviewed skill
- →Execution limits and result handoff for background tasks
A learned rule becomes persistent only after review, and scheduled tasks leave traceable results and boundaries.
Browser Projects Kit
Run ChatGPT and Claude projects with sources, files and fixed work states.
- →CURRENT-BRIEF.md, SOURCE-INDEX.md, DECISIONS.md, HANDOFF.md
- →Separate folders for intake, reviewed sources and results
- →A format to carry good results back into the vault
A new browser project can be started coherently from a few files and carried back into the vault.
NotebookLM Research Kit
Source-grounded research and study work without context chaos.
- →Curated notebooks by purpose instead of one bulk store
- →Per notebook: source map, guiding questions, desired output
- →Handoff format back into the vault
Each notebook has a clearly bounded source question and the research returns as an evidenced handoff.
Harness & Evaluation Kit
Make agentic flows measurable, safe and reproducible — rules become testable workflows.
- →Role cards with tool and zone rights
- →An eval set that grows with every failure case
- →Incident playbook and audit templates
A reference workflow can be run reproducibly, evaluated and stopped on a boundary violation.
Get the packages
ZIPStarter vault · work contracts (DE · EN · TR)ZIP · 42 files↓The starter vault is available right away as a ZIP — directly openable as an Obsidian vault, readable without special software. The other six packages follow in phases; each phase ends with its own download.
● Members only
Read the full article and download all files with a membership.
Unlock full article + downloads → Subscribe0 comments
● Loading comments…